1/* 2 * WMA compatible decoder 3 * Copyright (c) 2002 The FFmpeg Project 4 * 5 * This file is part of FFmpeg. 6 * 7 * FFmpeg is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU Lesser General Public 9 * License as published by the Free Software Foundation; either 10 * version 2.1 of the License, or (at your option) any later version. 11 * 12 * FFmpeg is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15 * Lesser General Public License for more details. 16 * 17 * You should have received a copy of the GNU Lesser General Public 18 * License along with FFmpeg; if not, write to the Free Software 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 20 */ 21 22/** 23 * @file 24 * WMA compatible decoder. 25 * This decoder handles Microsoft Windows Media Audio data, versions 1 & 2. 26 * WMA v1 is identified by audio format 0x160 in Microsoft media files 27 * (ASF/AVI/WAV). WMA v2 is identified by audio format 0x161. 28 * 29 * To use this decoder, a calling application must supply the extra data 30 * bytes provided with the WMA data. These are the extra, codec-specific 31 * bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes 32 * to the decoder using the extradata[_size] fields in AVCodecContext. There 33 * should be 4 extra bytes for v1 data and 6 extra bytes for v2 data. 34 */ 35 36#include "libavutil/attributes.h" 37#include "avcodec.h" 38#include "internal.h" 39#include "wma.h" 40 41#undef NDEBUG 42#include <assert.h> 43 44#define EXPVLCBITS 8 45#define EXPMAX ((19+EXPVLCBITS-1)/EXPVLCBITS) 46 47#define HGAINVLCBITS 9 48#define HGAINMAX ((13+HGAINVLCBITS-1)/HGAINVLCBITS) 49 50static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len); 51 52#ifdef TRACE 53static void dump_floats(WMACodecContext *s, const char *name, int prec, const float *tab, int n) 54{ 55 int i; 56 57 tprintf(s->avctx, "%s[%d]:\n", name, n); 58 for(i=0;i<n;i++) { 59 if ((i & 7) == 0) 60 tprintf(s->avctx, "%4d: ", i); 61 tprintf(s->avctx, " %8.*f", prec, tab[i]); 62 if ((i & 7) == 7) 63 tprintf(s->avctx, "\n"); 64 } 65 if ((i & 7) != 0) 66 tprintf(s->avctx, "\n"); 67} 68#endif 69 70static av_cold int wma_decode_init(AVCodecContext * avctx) 71{ 72 WMACodecContext *s = avctx->priv_data; 73 int i, flags2; 74 uint8_t *extradata; 75 76 if (!avctx->block_align) { 77 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n"); 78 return AVERROR(EINVAL); 79 } 80 81 s->avctx = avctx; 82 83 /* extract flag infos */ 84 flags2 = 0; 85 extradata = avctx->extradata; 86 if (avctx->codec->id == AV_CODEC_ID_WMAV1 && avctx->extradata_size >= 4) { 87 flags2 = AV_RL16(extradata+2); 88 } else if (avctx->codec->id == AV_CODEC_ID_WMAV2 && avctx->extradata_size >= 6) { 89 flags2 = AV_RL16(extradata+4); 90 } 91 92 s->use_exp_vlc = flags2 & 0x0001; 93 s->use_bit_reservoir = flags2 & 0x0002; 94 s->use_variable_block_len = flags2 & 0x0004; 95 96 if(avctx->codec->id == AV_CODEC_ID_WMAV2 && avctx->extradata_size >= 8){ 97 if(AV_RL16(extradata+4)==0xd && s->use_variable_block_len){ 98 av_log(avctx, AV_LOG_WARNING, "Disabling use_variable_block_len, if this fails contact the ffmpeg developers and send us the file\n"); 99 s->use_variable_block_len= 0; // this fixes issue1503 100 } 101 } 102 103 for (i=0; i<MAX_CHANNELS; i++) 104 s->max_exponent[i] = 1.0; 105 106 if(ff_wma_init(avctx, flags2)<0) 107 return -1; 108 109 /* init MDCT */ 110 for(i = 0; i < s->nb_block_sizes; i++) 111 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1, 1.0 / 32768.0); 112 113 if (s->use_noise_coding) { 114 init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits), 115 ff_wma_hgain_huffbits, 1, 1, 116 ff_wma_hgain_huffcodes, 2, 2, 0); 117 } 118 119 if (s->use_exp_vlc) { 120 init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_aac_scalefactor_bits), //FIXME move out of context 121 ff_aac_scalefactor_bits, 1, 1, 122 ff_aac_scalefactor_code, 4, 4, 0); 123 } else { 124 wma_lsp_to_curve_init(s, s->frame_len); 125 } 126 127 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; 128 129 return 0; 130} 131 132/** 133 * compute x^-0.25 with an exponent and mantissa table. We use linear 134 * interpolation to reduce the mantissa table size at a small speed 135 * expense (linear interpolation approximately doubles the number of 136 * bits of precision). 137 */ 138static inline float pow_m1_4(WMACodecContext *s, float x) 139{ 140 union { 141 float f; 142 unsigned int v; 143 } u, t; 144 unsigned int e, m; 145 float a, b; 146 147 u.f = x; 148 e = u.v >> 23; 149 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1); 150 /* build interpolation scale: 1 <= t < 2. */ 151 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23); 152 a = s->lsp_pow_m_table1[m]; 153 b = s->lsp_pow_m_table2[m]; 154 return s->lsp_pow_e_table[e] * (a + b * t.f); 155} 156 157static av_cold void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len) 158{ 159 float wdel, a, b; 160 int i, e, m; 161 162 wdel = M_PI / frame_len; 163 for(i=0;i<frame_len;i++) 164 s->lsp_cos_table[i] = 2.0f * cos(wdel * i); 165 166 /* tables for x^-0.25 computation */ 167 for(i=0;i<256;i++) { 168 e = i - 126; 169 s->lsp_pow_e_table[i] = pow(2.0, e * -0.25); 170 } 171 172 /* NOTE: these two tables are needed to avoid two operations in 173 pow_m1_4 */ 174 b = 1.0; 175 for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) { 176 m = (1 << LSP_POW_BITS) + i; 177 a = (float)m * (0.5 / (1 << LSP_POW_BITS)); 178 a = pow(a, -0.25); 179 s->lsp_pow_m_table1[i] = 2 * a - b; 180 s->lsp_pow_m_table2[i] = b - a; 181 b = a; 182 } 183} 184 185/** 186 * NOTE: We use the same code as Vorbis here 187 * @todo optimize it further with SSE/3Dnow 188 */ 189static void wma_lsp_to_curve(WMACodecContext *s, 190 float *out, float *val_max_ptr, 191 int n, float *lsp) 192{ 193 int i, j; 194 float p, q, w, v, val_max; 195 196 val_max = 0; 197 for(i=0;i<n;i++) { 198 p = 0.5f; 199 q = 0.5f; 200 w = s->lsp_cos_table[i]; 201 for(j=1;j<NB_LSP_COEFS;j+=2){ 202 q *= w - lsp[j - 1]; 203 p *= w - lsp[j]; 204 } 205 p *= p * (2.0f - w); 206 q *= q * (2.0f + w); 207 v = p + q; 208 v = pow_m1_4(s, v); 209 if (v > val_max) 210 val_max = v; 211 out[i] = v; 212 } 213 *val_max_ptr = val_max; 214} 215 216/** 217 * decode exponents coded with LSP coefficients (same idea as Vorbis) 218 */ 219static void decode_exp_lsp(WMACodecContext *s, int ch) 220{ 221 float lsp_coefs[NB_LSP_COEFS]; 222 int val, i; 223 224 for(i = 0; i < NB_LSP_COEFS; i++) { 225 if (i == 0 || i >= 8) 226 val = get_bits(&s->gb, 3); 227 else 228 val = get_bits(&s->gb, 4); 229 lsp_coefs[i] = ff_wma_lsp_codebook[i][val]; 230 } 231 232 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch], 233 s->block_len, lsp_coefs); 234} 235 236/** pow(10, i / 16.0) for i in -60..95 */ 237static const float pow_tab[] = { 238 1.7782794100389e-04, 2.0535250264571e-04, 239 2.3713737056617e-04, 2.7384196342644e-04, 240 3.1622776601684e-04, 3.6517412725484e-04, 241 4.2169650342858e-04, 4.8696752516586e-04, 242 5.6234132519035e-04, 6.4938163157621e-04, 243 7.4989420933246e-04, 8.6596432336006e-04, 244 1.0000000000000e-03, 1.1547819846895e-03, 245 1.3335214321633e-03, 1.5399265260595e-03, 246 1.7782794100389e-03, 2.0535250264571e-03, 247 2.3713737056617e-03, 2.7384196342644e-03, 248 3.1622776601684e-03, 3.6517412725484e-03, 249 4.2169650342858e-03, 4.8696752516586e-03, 250 5.6234132519035e-03, 6.4938163157621e-03, 251 7.4989420933246e-03, 8.6596432336006e-03, 252 1.0000000000000e-02, 1.1547819846895e-02, 253 1.3335214321633e-02, 1.5399265260595e-02, 254 1.7782794100389e-02, 2.0535250264571e-02, 255 2.3713737056617e-02, 2.7384196342644e-02, 256 3.1622776601684e-02, 3.6517412725484e-02, 257 4.2169650342858e-02, 4.8696752516586e-02, 258 5.6234132519035e-02, 6.4938163157621e-02, 259 7.4989420933246e-02, 8.6596432336007e-02, 260 1.0000000000000e-01, 1.1547819846895e-01, 261 1.3335214321633e-01, 1.5399265260595e-01, 262 1.7782794100389e-01, 2.0535250264571e-01, 263 2.3713737056617e-01, 2.7384196342644e-01, 264 3.1622776601684e-01, 3.6517412725484e-01, 265 4.2169650342858e-01, 4.8696752516586e-01, 266 5.6234132519035e-01, 6.4938163157621e-01, 267 7.4989420933246e-01, 8.6596432336007e-01, 268 1.0000000000000e+00, 1.1547819846895e+00, 269 1.3335214321633e+00, 1.5399265260595e+00, 270 1.7782794100389e+00, 2.0535250264571e+00, 271 2.3713737056617e+00, 2.7384196342644e+00, 272 3.1622776601684e+00, 3.6517412725484e+00, 273 4.2169650342858e+00, 4.8696752516586e+00, 274 5.6234132519035e+00, 6.4938163157621e+00, 275 7.4989420933246e+00, 8.6596432336007e+00, 276 1.0000000000000e+01, 1.1547819846895e+01, 277 1.3335214321633e+01, 1.5399265260595e+01, 278 1.7782794100389e+01, 2.0535250264571e+01, 279 2.3713737056617e+01, 2.7384196342644e+01, 280 3.1622776601684e+01, 3.6517412725484e+01, 281 4.2169650342858e+01, 4.8696752516586e+01, 282 5.6234132519035e+01, 6.4938163157621e+01, 283 7.4989420933246e+01, 8.6596432336007e+01, 284 1.0000000000000e+02, 1.1547819846895e+02, 285 1.3335214321633e+02, 1.5399265260595e+02, 286 1.7782794100389e+02, 2.0535250264571e+02, 287 2.3713737056617e+02, 2.7384196342644e+02, 288 3.1622776601684e+02, 3.6517412725484e+02, 289 4.2169650342858e+02, 4.8696752516586e+02, 290 5.6234132519035e+02, 6.4938163157621e+02, 291 7.4989420933246e+02, 8.6596432336007e+02, 292 1.0000000000000e+03, 1.1547819846895e+03, 293 1.3335214321633e+03, 1.5399265260595e+03, 294 1.7782794100389e+03, 2.0535250264571e+03, 295 2.3713737056617e+03, 2.7384196342644e+03, 296 3.1622776601684e+03, 3.6517412725484e+03, 297 4.2169650342858e+03, 4.8696752516586e+03, 298 5.6234132519035e+03, 6.4938163157621e+03, 299 7.4989420933246e+03, 8.6596432336007e+03, 300 1.0000000000000e+04, 1.1547819846895e+04, 301 1.3335214321633e+04, 1.5399265260595e+04, 302 1.7782794100389e+04, 2.0535250264571e+04, 303 2.3713737056617e+04, 2.7384196342644e+04, 304 3.1622776601684e+04, 3.6517412725484e+04, 305 4.2169650342858e+04, 4.8696752516586e+04, 306 5.6234132519035e+04, 6.4938163157621e+04, 307 7.4989420933246e+04, 8.6596432336007e+04, 308 1.0000000000000e+05, 1.1547819846895e+05, 309 1.3335214321633e+05, 1.5399265260595e+05, 310 1.7782794100389e+05, 2.0535250264571e+05, 311 2.3713737056617e+05, 2.7384196342644e+05, 312 3.1622776601684e+05, 3.6517412725484e+05, 313 4.2169650342858e+05, 4.8696752516586e+05, 314 5.6234132519035e+05, 6.4938163157621e+05, 315 7.4989420933246e+05, 8.6596432336007e+05, 316}; 317 318/** 319 * decode exponents coded with VLC codes 320 */ 321static int decode_exp_vlc(WMACodecContext *s, int ch) 322{ 323 int last_exp, n, code; 324 const uint16_t *ptr; 325 float v, max_scale; 326 uint32_t *q, *q_end, iv; 327 const float *ptab = pow_tab + 60; 328 const uint32_t *iptab = (const uint32_t*)ptab; 329 330 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; 331 q = (uint32_t *)s->exponents[ch]; 332 q_end = q + s->block_len; 333 max_scale = 0; 334 if (s->version == 1) { 335 last_exp = get_bits(&s->gb, 5) + 10; 336 v = ptab[last_exp]; 337 iv = iptab[last_exp]; 338 max_scale = v; 339 n = *ptr++; 340 switch (n & 3) do { 341 case 0: *q++ = iv; 342 case 3: *q++ = iv; 343 case 2: *q++ = iv; 344 case 1: *q++ = iv; 345 } while ((n -= 4) > 0); 346 }else 347 last_exp = 36; 348 349 while (q < q_end) { 350 code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX); 351 if (code < 0){ 352 av_log(s->avctx, AV_LOG_ERROR, "Exponent vlc invalid\n"); 353 return -1; 354 } 355 /* NOTE: this offset is the same as MPEG4 AAC ! */ 356 last_exp += code - 60; 357 if ((unsigned)last_exp + 60 >= FF_ARRAY_ELEMS(pow_tab)) { 358 av_log(s->avctx, AV_LOG_ERROR, "Exponent out of range: %d\n", 359 last_exp); 360 return -1; 361 } 362 v = ptab[last_exp]; 363 iv = iptab[last_exp]; 364 if (v > max_scale) 365 max_scale = v; 366 n = *ptr++; 367 switch (n & 3) do { 368 case 0: *q++ = iv; 369 case 3: *q++ = iv; 370 case 2: *q++ = iv; 371 case 1: *q++ = iv; 372 } while ((n -= 4) > 0); 373 } 374 s->max_exponent[ch] = max_scale; 375 return 0; 376} 377 378 379/** 380 * Apply MDCT window and add into output. 381 * 382 * We ensure that when the windows overlap their squared sum 383 * is always 1 (MDCT reconstruction rule). 384 */ 385static void wma_window(WMACodecContext *s, float *out) 386{ 387 float *in = s->output; 388 int block_len, bsize, n; 389 390 /* left part */ 391 if (s->block_len_bits <= s->prev_block_len_bits) { 392 block_len = s->block_len; 393 bsize = s->frame_len_bits - s->block_len_bits; 394 395 s->fdsp.vector_fmul_add(out, in, s->windows[bsize], 396 out, block_len); 397 398 } else { 399 block_len = 1 << s->prev_block_len_bits; 400 n = (s->block_len - block_len) / 2; 401 bsize = s->frame_len_bits - s->prev_block_len_bits; 402 403 s->fdsp.vector_fmul_add(out+n, in+n, s->windows[bsize], 404 out+n, block_len); 405 406 memcpy(out+n+block_len, in+n+block_len, n*sizeof(float)); 407 } 408 409 out += s->block_len; 410 in += s->block_len; 411 412 /* right part */ 413 if (s->block_len_bits <= s->next_block_len_bits) { 414 block_len = s->block_len; 415 bsize = s->frame_len_bits - s->block_len_bits; 416 417 s->fdsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len); 418 419 } else { 420 block_len = 1 << s->next_block_len_bits; 421 n = (s->block_len - block_len) / 2; 422 bsize = s->frame_len_bits - s->next_block_len_bits; 423 424 memcpy(out, in, n*sizeof(float)); 425 426 s->fdsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len); 427 428 memset(out+n+block_len, 0, n*sizeof(float)); 429 } 430} 431 432 433/** 434 * @return 0 if OK. 1 if last block of frame. return -1 if 435 * unrecorrable error. 436 */ 437static int wma_decode_block(WMACodecContext *s) 438{ 439 int n, v, a, ch, bsize; 440 int coef_nb_bits, total_gain; 441 int nb_coefs[MAX_CHANNELS]; 442 float mdct_norm; 443 FFTContext *mdct; 444 445#ifdef TRACE 446 tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num); 447#endif 448 449 /* compute current block length */ 450 if (s->use_variable_block_len) { 451 n = av_log2(s->nb_block_sizes - 1) + 1; 452 453 if (s->reset_block_lengths) { 454 s->reset_block_lengths = 0; 455 v = get_bits(&s->gb, n); 456 if (v >= s->nb_block_sizes){ 457 av_log(s->avctx, AV_LOG_ERROR, "prev_block_len_bits %d out of range\n", s->frame_len_bits - v); 458 return -1; 459 } 460 s->prev_block_len_bits = s->frame_len_bits - v; 461 v = get_bits(&s->gb, n); 462 if (v >= s->nb_block_sizes){ 463 av_log(s->avctx, AV_LOG_ERROR, "block_len_bits %d out of range\n", s->frame_len_bits - v); 464 return -1; 465 } 466 s->block_len_bits = s->frame_len_bits - v; 467 } else { 468 /* update block lengths */ 469 s->prev_block_len_bits = s->block_len_bits; 470 s->block_len_bits = s->next_block_len_bits; 471 } 472 v = get_bits(&s->gb, n); 473 if (v >= s->nb_block_sizes){ 474 av_log(s->avctx, AV_LOG_ERROR, "next_block_len_bits %d out of range\n", s->frame_len_bits - v); 475 return -1; 476 } 477 s->next_block_len_bits = s->frame_len_bits - v; 478 } else { 479 /* fixed block len */ 480 s->next_block_len_bits = s->frame_len_bits; 481 s->prev_block_len_bits = s->frame_len_bits; 482 s->block_len_bits = s->frame_len_bits; 483 } 484 485 if (s->frame_len_bits - s->block_len_bits >= s->nb_block_sizes){ 486 av_log(s->avctx, AV_LOG_ERROR, "block_len_bits not initialized to a valid value\n"); 487 return -1; 488 } 489 490 /* now check if the block length is coherent with the frame length */ 491 s->block_len = 1 << s->block_len_bits; 492 if ((s->block_pos + s->block_len) > s->frame_len){ 493 av_log(s->avctx, AV_LOG_ERROR, "frame_len overflow\n"); 494 return -1; 495 } 496 497 if (s->avctx->channels == 2) { 498 s->ms_stereo = get_bits1(&s->gb); 499 } 500 v = 0; 501 for(ch = 0; ch < s->avctx->channels; ch++) { 502 a = get_bits1(&s->gb); 503 s->channel_coded[ch] = a; 504 v |= a; 505 } 506 507 bsize = s->frame_len_bits - s->block_len_bits; 508 509 /* if no channel coded, no need to go further */ 510 /* XXX: fix potential framing problems */ 511 if (!v) 512 goto next; 513 514 /* read total gain and extract corresponding number of bits for 515 coef escape coding */ 516 total_gain = 1; 517 for(;;) { 518 if (get_bits_left(&s->gb) < 7) { 519 av_log(s->avctx, AV_LOG_ERROR, "total_gain overread\n"); 520 return AVERROR_INVALIDDATA; 521 } 522 a = get_bits(&s->gb, 7); 523 total_gain += a; 524 if (a != 127) 525 break; 526 } 527 528 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain); 529 530 /* compute number of coefficients */ 531 n = s->coefs_end[bsize] - s->coefs_start; 532 for(ch = 0; ch < s->avctx->channels; ch++) 533 nb_coefs[ch] = n; 534 535 /* complex coding */ 536 if (s->use_noise_coding) { 537 538 for(ch = 0; ch < s->avctx->channels; ch++) { 539 if (s->channel_coded[ch]) { 540 int i, n, a; 541 n = s->exponent_high_sizes[bsize]; 542 for(i=0;i<n;i++) { 543 a = get_bits1(&s->gb); 544 s->high_band_coded[ch][i] = a; 545 /* if noise coding, the coefficients are not transmitted */ 546 if (a) 547 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; 548 } 549 } 550 } 551 for(ch = 0; ch < s->avctx->channels; ch++) { 552 if (s->channel_coded[ch]) { 553 int i, n, val, code; 554 555 n = s->exponent_high_sizes[bsize]; 556 val = (int)0x80000000; 557 for(i=0;i<n;i++) { 558 if (s->high_band_coded[ch][i]) { 559 if (val == (int)0x80000000) { 560 val = get_bits(&s->gb, 7) - 19; 561 } else { 562 code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX); 563 if (code < 0){ 564 av_log(s->avctx, AV_LOG_ERROR, "hgain vlc invalid\n"); 565 return -1; 566 } 567 val += code - 18; 568 } 569 s->high_band_values[ch][i] = val; 570 } 571 } 572 } 573 } 574 } 575 576 /* exponents can be reused in short blocks. */ 577 if ((s->block_len_bits == s->frame_len_bits) || 578 get_bits1(&s->gb)) { 579 for(ch = 0; ch < s->avctx->channels; ch++) { 580 if (s->channel_coded[ch]) { 581 if (s->use_exp_vlc) { 582 if (decode_exp_vlc(s, ch) < 0) 583 return -1; 584 } else { 585 decode_exp_lsp(s, ch); 586 } 587 s->exponents_bsize[ch] = bsize; 588 } 589 } 590 } 591 592 /* parse spectral coefficients : just RLE encoding */ 593 for (ch = 0; ch < s->avctx->channels; ch++) { 594 if (s->channel_coded[ch]) { 595 int tindex; 596 WMACoef* ptr = &s->coefs1[ch][0]; 597 598 /* special VLC tables are used for ms stereo because 599 there is potentially less energy there */ 600 tindex = (ch == 1 && s->ms_stereo); 601 memset(ptr, 0, s->block_len * sizeof(WMACoef)); 602 ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex], 603 s->level_table[tindex], s->run_table[tindex], 604 0, ptr, 0, nb_coefs[ch], 605 s->block_len, s->frame_len_bits, coef_nb_bits); 606 } 607 if (s->version == 1 && s->avctx->channels >= 2) { 608 align_get_bits(&s->gb); 609 } 610 } 611 612 /* normalize */ 613 { 614 int n4 = s->block_len / 2; 615 mdct_norm = 1.0 / (float)n4; 616 if (s->version == 1) { 617 mdct_norm *= sqrt(n4); 618 } 619 } 620 621 /* finally compute the MDCT coefficients */ 622 for (ch = 0; ch < s->avctx->channels; ch++) { 623 if (s->channel_coded[ch]) { 624 WMACoef *coefs1; 625 float *coefs, *exponents, mult, mult1, noise; 626 int i, j, n, n1, last_high_band, esize; 627 float exp_power[HIGH_BAND_MAX_SIZE]; 628 629 coefs1 = s->coefs1[ch]; 630 exponents = s->exponents[ch]; 631 esize = s->exponents_bsize[ch]; 632 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; 633 mult *= mdct_norm; 634 coefs = s->coefs[ch]; 635 if (s->use_noise_coding) { 636 mult1 = mult; 637 /* very low freqs : noise */ 638 for(i = 0;i < s->coefs_start; i++) { 639 *coefs++ = s->noise_table[s->noise_index] * 640 exponents[i<<bsize>>esize] * mult1; 641 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 642 } 643 644 n1 = s->exponent_high_sizes[bsize]; 645 646 /* compute power of high bands */ 647 exponents = s->exponents[ch] + 648 (s->high_band_start[bsize]<<bsize>>esize); 649 last_high_band = 0; /* avoid warning */ 650 for(j=0;j<n1;j++) { 651 n = s->exponent_high_bands[s->frame_len_bits - 652 s->block_len_bits][j]; 653 if (s->high_band_coded[ch][j]) { 654 float e2, v; 655 e2 = 0; 656 for(i = 0;i < n; i++) { 657 v = exponents[i<<bsize>>esize]; 658 e2 += v * v; 659 } 660 exp_power[j] = e2 / n; 661 last_high_band = j; 662 tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n); 663 } 664 exponents += n<<bsize>>esize; 665 } 666 667 /* main freqs and high freqs */ 668 exponents = s->exponents[ch] + (s->coefs_start<<bsize>>esize); 669 for(j=-1;j<n1;j++) { 670 if (j < 0) { 671 n = s->high_band_start[bsize] - 672 s->coefs_start; 673 } else { 674 n = s->exponent_high_bands[s->frame_len_bits - 675 s->block_len_bits][j]; 676 } 677 if (j >= 0 && s->high_band_coded[ch][j]) { 678 /* use noise with specified power */ 679 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]); 680 /* XXX: use a table */ 681 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05); 682 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult); 683 mult1 *= mdct_norm; 684 for(i = 0;i < n; i++) { 685 noise = s->noise_table[s->noise_index]; 686 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 687 *coefs++ = noise * 688 exponents[i<<bsize>>esize] * mult1; 689 } 690 exponents += n<<bsize>>esize; 691 } else { 692 /* coded values + small noise */ 693 for(i = 0;i < n; i++) { 694 noise = s->noise_table[s->noise_index]; 695 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 696 *coefs++ = ((*coefs1++) + noise) * 697 exponents[i<<bsize>>esize] * mult; 698 } 699 exponents += n<<bsize>>esize; 700 } 701 } 702 703 /* very high freqs : noise */ 704 n = s->block_len - s->coefs_end[bsize]; 705 mult1 = mult * exponents[((-1<<bsize))>>esize]; 706 for(i = 0; i < n; i++) { 707 *coefs++ = s->noise_table[s->noise_index] * mult1; 708 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 709 } 710 } else { 711 /* XXX: optimize more */ 712 for(i = 0;i < s->coefs_start; i++) 713 *coefs++ = 0.0; 714 n = nb_coefs[ch]; 715 for(i = 0;i < n; i++) { 716 *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult; 717 } 718 n = s->block_len - s->coefs_end[bsize]; 719 for(i = 0;i < n; i++) 720 *coefs++ = 0.0; 721 } 722 } 723 } 724 725#ifdef TRACE 726 for (ch = 0; ch < s->avctx->channels; ch++) { 727 if (s->channel_coded[ch]) { 728 dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len); 729 dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len); 730 } 731 } 732#endif 733 734 if (s->ms_stereo && s->channel_coded[1]) { 735 /* nominal case for ms stereo: we do it before mdct */ 736 /* no need to optimize this case because it should almost 737 never happen */ 738 if (!s->channel_coded[0]) { 739 tprintf(s->avctx, "rare ms-stereo case happened\n"); 740 memset(s->coefs[0], 0, sizeof(float) * s->block_len); 741 s->channel_coded[0] = 1; 742 } 743 744 s->fdsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len); 745 } 746 747next: 748 mdct = &s->mdct_ctx[bsize]; 749 750 for (ch = 0; ch < s->avctx->channels; ch++) { 751 int n4, index; 752 753 n4 = s->block_len / 2; 754 if(s->channel_coded[ch]){ 755 mdct->imdct_calc(mdct, s->output, s->coefs[ch]); 756 }else if(!(s->ms_stereo && ch==1)) 757 memset(s->output, 0, sizeof(s->output)); 758 759 /* multiply by the window and add in the frame */ 760 index = (s->frame_len / 2) + s->block_pos - n4; 761 wma_window(s, &s->frame_out[ch][index]); 762 } 763 764 /* update block number */ 765 s->block_num++; 766 s->block_pos += s->block_len; 767 if (s->block_pos >= s->frame_len) 768 return 1; 769 else 770 return 0; 771} 772 773/* decode a frame of frame_len samples */ 774static int wma_decode_frame(WMACodecContext *s, float **samples, 775 int samples_offset) 776{ 777 int ret, ch; 778 779#ifdef TRACE 780 tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len); 781#endif 782 783 /* read each block */ 784 s->block_num = 0; 785 s->block_pos = 0; 786 for(;;) { 787 ret = wma_decode_block(s); 788 if (ret < 0) 789 return -1; 790 if (ret) 791 break; 792 } 793 794 for (ch = 0; ch < s->avctx->channels; ch++) { 795 /* copy current block to output */ 796 memcpy(samples[ch] + samples_offset, s->frame_out[ch], 797 s->frame_len * sizeof(*s->frame_out[ch])); 798 /* prepare for next block */ 799 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len], 800 s->frame_len * sizeof(*s->frame_out[ch])); 801 802#ifdef TRACE 803 dump_floats(s, "samples", 6, samples[ch] + samples_offset, s->frame_len); 804#endif 805 } 806 807 return 0; 808} 809 810static int wma_decode_superframe(AVCodecContext *avctx, void *data, 811 int *got_frame_ptr, AVPacket *avpkt) 812{ 813 AVFrame *frame = data; 814 const uint8_t *buf = avpkt->data; 815 int buf_size = avpkt->size; 816 WMACodecContext *s = avctx->priv_data; 817 int nb_frames, bit_offset, i, pos, len, ret; 818 uint8_t *q; 819 float **samples; 820 int samples_offset; 821 822 tprintf(avctx, "***decode_superframe:\n"); 823 824 if(buf_size==0){ 825 s->last_superframe_len = 0; 826 return 0; 827 } 828 if (buf_size < avctx->block_align) { 829 av_log(avctx, AV_LOG_ERROR, 830 "Input packet size too small (%d < %d)\n", 831 buf_size, avctx->block_align); 832 return AVERROR_INVALIDDATA; 833 } 834 if(avctx->block_align) 835 buf_size = avctx->block_align; 836 837 init_get_bits(&s->gb, buf, buf_size*8); 838 839 if (s->use_bit_reservoir) { 840 /* read super frame header */ 841 skip_bits(&s->gb, 4); /* super frame index */ 842 nb_frames = get_bits(&s->gb, 4) - (s->last_superframe_len <= 0); 843 if (nb_frames <= 0) { 844 av_log(avctx, AV_LOG_ERROR, "nb_frames is %d\n", nb_frames); 845 return AVERROR_INVALIDDATA; 846 } 847 } else { 848 nb_frames = 1; 849 } 850 851 /* get output buffer */ 852 frame->nb_samples = nb_frames * s->frame_len; 853 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) 854 return ret; 855 samples = (float **)frame->extended_data; 856 samples_offset = 0; 857 858 if (s->use_bit_reservoir) { 859 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3); 860 if (bit_offset > get_bits_left(&s->gb)) { 861 av_log(avctx, AV_LOG_ERROR, 862 "Invalid last frame bit offset %d > buf size %d (%d)\n", 863 bit_offset, get_bits_left(&s->gb), buf_size); 864 goto fail; 865 } 866 867 if (s->last_superframe_len > 0) { 868 /* add bit_offset bits to last frame */ 869 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) > 870 MAX_CODED_SUPERFRAME_SIZE) 871 goto fail; 872 q = s->last_superframe + s->last_superframe_len; 873 len = bit_offset; 874 while (len > 7) { 875 *q++ = (get_bits)(&s->gb, 8); 876 len -= 8; 877 } 878 if (len > 0) { 879 *q++ = (get_bits)(&s->gb, len) << (8 - len); 880 } 881 memset(q, 0, FF_INPUT_BUFFER_PADDING_SIZE); 882 883 /* XXX: bit_offset bits into last frame */ 884 init_get_bits(&s->gb, s->last_superframe, s->last_superframe_len * 8 + bit_offset); 885 /* skip unused bits */ 886 if (s->last_bitoffset > 0) 887 skip_bits(&s->gb, s->last_bitoffset); 888 /* this frame is stored in the last superframe and in the 889 current one */ 890 if (wma_decode_frame(s, samples, samples_offset) < 0) 891 goto fail; 892 samples_offset += s->frame_len; 893 nb_frames--; 894 } 895 896 /* read each frame starting from bit_offset */ 897 pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3; 898 if (pos >= MAX_CODED_SUPERFRAME_SIZE * 8 || pos > buf_size * 8) 899 return AVERROR_INVALIDDATA; 900 init_get_bits(&s->gb, buf + (pos >> 3), (buf_size - (pos >> 3))*8); 901 len = pos & 7; 902 if (len > 0) 903 skip_bits(&s->gb, len); 904 905 s->reset_block_lengths = 1; 906 for(i=0;i<nb_frames;i++) { 907 if (wma_decode_frame(s, samples, samples_offset) < 0) 908 goto fail; 909 samples_offset += s->frame_len; 910 } 911 912 /* we copy the end of the frame in the last frame buffer */ 913 pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7); 914 s->last_bitoffset = pos & 7; 915 pos >>= 3; 916 len = buf_size - pos; 917 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) { 918 av_log(s->avctx, AV_LOG_ERROR, "len %d invalid\n", len); 919 goto fail; 920 } 921 s->last_superframe_len = len; 922 memcpy(s->last_superframe, buf + pos, len); 923 } else { 924 /* single frame decode */ 925 if (wma_decode_frame(s, samples, samples_offset) < 0) 926 goto fail; 927 samples_offset += s->frame_len; 928 } 929 930 av_dlog(s->avctx, "%d %d %d %d outbytes:%"PTRDIFF_SPECIFIER" eaten:%d\n", 931 s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len, 932 (int8_t *)samples - (int8_t *)data, avctx->block_align); 933 934 *got_frame_ptr = 1; 935 936 return buf_size; 937 fail: 938 /* when error, we reset the bit reservoir */ 939 s->last_superframe_len = 0; 940 return -1; 941} 942 943static av_cold void flush(AVCodecContext *avctx) 944{ 945 WMACodecContext *s = avctx->priv_data; 946 947 s->last_bitoffset= 948 s->last_superframe_len= 0; 949} 950 951#if CONFIG_WMAV1_DECODER 952AVCodec ff_wmav1_decoder = { 953 .name = "wmav1", 954 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"), 955 .type = AVMEDIA_TYPE_AUDIO, 956 .id = AV_CODEC_ID_WMAV1, 957 .priv_data_size = sizeof(WMACodecContext), 958 .init = wma_decode_init, 959 .close = ff_wma_end, 960 .decode = wma_decode_superframe, 961 .flush = flush, 962 .capabilities = CODEC_CAP_DR1, 963 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, 964 AV_SAMPLE_FMT_NONE }, 965}; 966#endif 967#if CONFIG_WMAV2_DECODER 968AVCodec ff_wmav2_decoder = { 969 .name = "wmav2", 970 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"), 971 .type = AVMEDIA_TYPE_AUDIO, 972 .id = AV_CODEC_ID_WMAV2, 973 .priv_data_size = sizeof(WMACodecContext), 974 .init = wma_decode_init, 975 .close = ff_wma_end, 976 .decode = wma_decode_superframe, 977 .flush = flush, 978 .capabilities = CODEC_CAP_DR1, 979 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, 980 AV_SAMPLE_FMT_NONE }, 981}; 982#endif 983